OCR Based on ARG Matching Algorithm
Veljko D. Papic, Zeljko M. Djurovic, Branko D. Kovačević · 2006
The problem of alphabet character recognition, using known ARG (attributed relational graph) based technique, is considered. The object (character) representation, which is invariant to arbitrary projective transformation, is based on unary and binary measures. Transformation matrices for all transformation types (translation, similarity, affine and perspective transform) are defined. Basis matrices are obtained and barycentric coordinates, unary and binary measurements are defined. It is experimentally confirmed that the binary measures are invariant related to projective transform and so may be used as features for character recognition.